#include <iostream>

#include <boost/compute/core.hpp>
#include <boost/compute/algorithm/copy.hpp>
#include <boost/compute/container/vector.hpp>
#include <boost/compute/utility/source.hpp>

#include <boost/compute/system.hpp>



namespace compute = boost::compute;

// this example demonstrates how to use the Boost.Compute classes to
// setup and run a simple vector addition kernel on the GPU
int main()
{
    // get the default device
    compute::device device = compute::system::default_device();

    // create a context for the device
    compute::context context(device);

    // setup input arrays
    std::vector<float> a = { 1, 2, 3, 4 };
    std::vector<float> b = { 5, 6, 7, 8 };

    // make space for the output
    std::vector<float> c = { 5, 6, 7, 8 };

    // create memory buffers for the input and output
    boost::compute::vector<float> a_d(4 , context);
    boost::compute::vector<float> b_d(4, context);
    boost::compute::vector<float> c_d(4, context);

    // source code for the add kernel
    const char source[] =
        "__kernel void add(__global const float *a,"
        "                  __global const float *b,"
        "                  __global float *c)"
        "{"
        "    const uint i = get_global_id(0);"
        "    c[i] = a[i] + b[i];"
        "}";

    // create the program with the source
    compute::program program =
        compute::program::create_with_source(source, context);
    // compile the program
    program.build();

    // create the kernel
    compute::kernel kernel(program, "add");

    // create a command queue
    compute::command_queue queue(context, device);

    compute::copy(a.begin(), a.begin() + 3, a_d.begin(), queue);
    compute::copy(b.begin(), b.begin() + 3, b_d.begin(), queue);
    compute::copy(c.begin(), c.begin() + 3, c_d.begin(), queue);


    // set the kernel arguments
    kernel.set_arg(0, a_d);
    kernel.set_arg(1, a_d);
    kernel.set_arg(2, a_d);


    // run the add kernel
    queue.enqueue_1d_range_kernel(kernel, 0, 4, 0);
    // transfer results back to the host array 'c'
    // compute::copy(c_d.begin(), c_d.begin() + 3, c.begin(), queue);

    // print out results in 'c'
    std::cout << "c: [" << c[0] << ", "
                        << c[1] << ", "
                        << c[2] << ", "
                        << c[3] << "]" << std::endl;

    return 0;
}